5 Prescribing the penetration of geodesic lines
5.1 Climbing in balls and horoballs
In this subsection, we construct geodesic rays or lines having prescribed penetration properties in a ball or a horoball, while essentially avoiding a family of almost disjoint convex subsets. Let us consider the penetration height and inner projection penetration maps first in horoballs and then in balls. Note that in these cases, if f0 = phC0, then kf0−phC0k∞=0 and if f0=ippC0, then kf0−phC0k∞≤c′1(∞) by Section3.1.
Theorem 5.1 Letǫ ∈R∗+∪ {∞}, δ, κ ≥ 0; let X be a complete simply connected Riemannian manifold with sectional curvature at most −1 and dimension at least 3;
let ξ0 ∈ X∪∂∞X; let C0 be a horoball such that ξ0 ∈/ C0∪∂∞C0; let f0 = phC0
or f0 = ippC0; let (Cn)n∈N−{0} be a family of ǫ-convex subsets of X ; for every n∈ N− {0} such that ξ0 ∈/ Cn∪∂∞Cn, let fn : Tξ1
0X → [0,+∞] be a κ-penetration map in Cn. If diam(Cn∩Cm) ≤ δ for all n,m in N with n 6= m, then, for every h≥c′′1(ǫ, δ,kf0−phC0k∞), there exists a geodesic ray or line γ starting from ξ0 and entering C0at time 0, such that f0(γ)=h and fn(γ)≤c′′1(ǫ, δ,kf0−phC0k∞)+κ for every n≥1such that γ ]δ,+∞[
meets Cn.
Proof. Let h≥c′′1 =c′′1(ǫ, δ,kf0−phC0k∞). In order to apply Proposition4.10, define ǫ0 =ǫ, δ0=δ, κ0 =c′1(∞)=2 log(1+√
2) and h′0=h0(ǫ0, δ0, κ0). Recall thatphC0 and ippC0 are κ0-penetration maps for C0 by Lemma3.3. For every n ∈ N− {0} such thatξ0 ∈/ Cn∪∂∞Cn, let us apply Proposition3.7Case (1) to C=C0, C′ =Cn, f =f0, f′ =ℓCn, h′ =h′0, so that hmin=2 c′1(ǫ)+2δ+kf0−phC0k∞ and hmin0 =2δ. Note that hmin0 ≤h′0, as
h′0≥h′(ǫ,sinh(δ+c1))≥2 sinh(δ+c1)≥2δ ,
by the definitions of h0 and of h′(·,·). As h ≥ c′′1 ≥ hmin by the definition of c′′1, Proposition 3.7 Case (1) hence implies that (Cn)n∈N satisfies the Local prescription property (iv). Thus by Proposition4.10, there exists a geodesic ray or line γ starting at ξ0 such that f0(γ) = h and ℓCn(γ) ≤ h′1(ǫ0, δ0,h0(ǫ0, δ0, κ0)), which implies that fn(γ)≤c′′1 +κ, for every n≥1 such thatγ(]δ,+∞[) meets Cn. The proof of the corresponding result when C0 is a ball of radius R≥ǫis the same, using Case (2) of Proposition 3.7 instead of Case (1). This requires h ≤ hmax = 2R−2c′1(ǫ)− kf0−phC0k∞. To be nonempty, the following result requires
R≥c′′1(ǫ, δ,kf0−phC0k∞)/2+ c′1(ǫ)+kf0−phC0k∞/2.
Theorem 5.2 Letǫ >0,δ, κ≥0; let X be a complete simply connected Riemannian manifold with sectional curvature at most−1and dimension at least 3; let C0be a ball of radius R≥ǫ; letξ0 ∈(X∪∂∞X)−C0; let f0=phC0 or f0=ippC0; let (Cn)n∈N−{0} be a family ofǫ-convex subsets of X ; for every n∈N−{0}such thatξ0∈/Cn∪∂∞Cn, let fn : Tξ1
0X →[0,+∞] be aκ-penetration map in Cn. If diam(Cn∩Cm)≤δ for all n,minNwith n6=m, then, for every
h∈h
c′′1 ǫ, δ,kf0−phC0k∞
, 2R−2c′1(ǫ)− kf0−phC0k∞i ,
there exists a geodesic ray or line γ starting from ξ0 and entering C0 at time 0, such that f0(γ) = h and fn(γ) ≤ c′′1 ǫ, δ,kf0−phC0k∞
+κ for every n ≥ 1 such that
γ(]δ,+∞[)meets Cn.
Varying the family (Cn)n∈N−{0} of ǫ-convex subsets appearing in Theorems 5.1 and 5.2, among balls of radius at least ǫ, horoballs, ǫ-neighbourhoods of totally geodesic subspaces, etc, we get several corollaries. We will only state two of them, Corollaries 5.3and 5.5, which have applications to equivariant families. The proofs of these re-sults are simplified versions of the proof of Corollary4.11, giving better (though very probably not optimal) constants.
Corollary 5.3 Let X be a complete simply connected Riemannian manifold with sec-tional curvature at most−1 and dimension at least 3, and let Hn
n∈N be a family of horoballs in X with pairwise disjoint interiors. Then, for every h ≥ c′′1(∞,0,0) ≈ 6.5032, there exists a geodesic line γ′ such that phH0(γ′) = h and phHn(γ′) ≤ c′′2(∞)≈8.3712 for every n≥1.
Proof. Let c′′1 = c′′1(∞,0,0) and c′′2 =c′′2(∞). Let C0 = H0 and let ξ be a point in
∂∞X−∂∞C0. We apply Theorem5.1with ǫ=∞,δ =0, κ=0, ξ0 =ξ, Cn=Hn for every n inN, f0 =phC0, and fn=ℓCn for every n6=0 such thatξ0∈/ Cn∪∂∞Cn. Note that for every n∈N− {0}, the map fnis aκ-penetration map in Cn. As h≥c′′1, there exists a geodesic line γ starting from ξ and entering C0 at time 0, such that phC0(γ) = h and ℓCn(γ) ≤ c′′1 for every n ∈ N− {0} such that γ meets Cn at a positive time.
Letξ′ be the other endpoint ofγ. This point is not in ∂∞C0. Applying Theorem5.1 again, as above except that now ξ0 = ξ′, we get that there exists a geodesic line γ′ starting from ξ′ and entering C0 at time 0, such that phC0(γ′) =h andℓCn(γ′) ≤c′′1 for every n∈N− {0} such thatγ′ meets Cn at a positive time.
Assume by absurd that there exists n∈N− {0} such that phCn(γ′) >c′′2 >0. Then γ′ enters Cn at the point x′n, exiting it at the point y′nat a nonpositive time, as c′′2 >c′′1. In particular, if x′ = γ′(0) is the entering point of γ′ in C0, then x′,y′n,x′n, ξ′ are in this order onγ′ (see the picture in the proof of Corollary4.11).
Let y be the exiting point ofγ out of H0. With c1 =1/19, as in the proof of Lemma 4.6, sincephC0(γ) andphC0(γ′) are equal to
h≥c′′1 ≥h′1(∞,0,h0(∞,0,c′1(∞)))≥h0(∞,0,c′1(∞))=h′(∞,sinh c1) by the definition of c′′1,h′1,h0, we have d(x′,y)≤c1.
Letξnbe the point at infinity of Hn. Let p′ be the point in [x′n,y′n] the closest to ξn, so that
d(p′,y′n)≥βξn(y′n,p′)=phCn(γ′)/2>c′′2/2.
Let p be the point of γ the closest to p′. By convexity and the definition of c′′2, we have
d(p′,p)=d(p′, γ) ≤d(x′, γ)≤d(x′,y)≤c1<c′′2/2.
Hence p belongs to the interior of Cn. If p ∈]ξ,y], then the closest point to x′ on γ lies in ]ξ,p[ and by convexity,
c′′2/2<d(p′,y′n)≤d(p′,x′)≤d(p′,p)+d(p,x′)≤d(p′,p)+d(y,x′)≤2c1, a contradiction, as by the definition of c′′2, of c′′1 and of h′(ǫ, η) (see the equations (- 5 -) and (- 11 -)), we have
c′′2 ≥c′′1+2 c1≥h′(ǫ,sinh c1)+2 c1 ≥2 sinh c1+2c1 >4 c1 .
Hence p ∈ ]y, ξ′[ ⊂ γ(]0,+∞[), so that γ meets Cn at a positive time. But, by Lemma3.3and the definition of c′′2,
ℓCn(γ)≥phCn(γ)−c′1(∞) ≥2βξn(y′n,p)−c′1(∞)
≥2(βξn(y′n,p′)−d(p,p′))−c′1(∞)>2(c′′2/2−c1)−c′1(∞)=c′′1 .
This contradicts the construction ofγ.
Let M be a complete nonelementary geometrically finite Riemannian manifold with sectional curvature at most−1 (see for instance [Bow] for a general reference). Recall that a cusp of M is an asymptotic class of minimizing geodesic rays in M along which the injectivity radius converges to 0. If M has finite volume, then the set of cusps of M is in bijection with the (finite) set of ends of M , by the map which associates to a representative of a cusp the end of M towards which it converges. Let π : Me → M be a universal Riemannian covering of M , with covering group Γ. If e is a cusp of M , andρe a minimizing geodesic ray in the class e, as M is geometrically finite and nonelementary, there exists a horoball He inM centered at the point at infinitye ξe of a fixed liftρee ofρe inM , such thate γHe and He have disjoint interiors ifγ ∈Γdoes not fix ξe (see for instance [BK,Bow,HP5]). This horoball is unique if maximal (for the inclusion). The image Ve of He in M is called a Margulis neighbourhood of e, and the maximal Margulis neighbourhood of e if He is maximal. In this Section5, we assume that He is maximal, and thatρee starts from the boundary of He. Let hte : M →R be the map defined by
hte(x)= lim
t→∞(t−d(ρe(t),x)),
called the height function with respect to e. Let maxhte : T1M→Rbe defined by maxhte(γ)=sup
t∈R
hte(γ(t)).
The maximum height spectrum of the pair (M,e) is the subset of ]− ∞,+∞] defined by
MaxSp(M,e)=maxhte(T1M).
Corollary 5.4 Let M be a complete, nonelementary geometrically finite Riemannian manifold with sectional curvature at most−1and dimension at least 3, and let e be a cusp of M . Then MaxSp(M,e)contains [c′′2(∞)/2,+∞].
Note that c′′2(∞)/2≈4.1856, hence Theorem1.2of the introduction follows.
Proof. With the above notations, let (Hn)n∈N be the Γ-equivariant family of horoballs in M with pairwise disjoint interiors such that He 0 = He. Apply Corollary 5.3 to this family to get, for every h ≥ c′′2(∞) ≥ c′′1(∞,0,0), a geodesic line γe in M withe phH0(γe) = h and phHn(eγ) ≤ c′′2(∞) for every n ≥ 1. Let γ be the locally geodesic line in M image by π of eγ. Observe that hte◦π = βHn in Hn and that phHn(eγ) = 2 supt∈RβHn(eγ(t)) (see Section 3.1). Hence supt∈Rhte(γ(t)) = h/2 and the result
follows.
Schmidt and Sheingorn [SS] treated the case of two-dimensional manifolds of con-stant curvature −1 (hyperbolic surfaces) with a cusp. They showed that in that case MaxSp(M,e) contains the interval [log 100,+∞] ≈ [4.61,+∞]. This paper [SS]
was a starting point of our investigations, although the method we use is quite differ-ent from theirs.
Let P be a (necessarily finite) nonempty set of cusps of M . For every e in P, choose a maximal horoball He, with point at infinity ξe as above Corollary 5.4. The horoballs of the family (gHe)g∈Γ/Γξe,e∈P may have intersecting interiors. But as M is geometrically finite and nonelementary, there exists (see [BK,Bow]) t ≥ 0 such that two distinct elements in (gHe[t])g∈Γ/Γξ
e,e∈P have disjoint interiors. Let tP be the lower bound of all such t’s. For everyγ ∈T1M , define
maxhtP(γ) =max
e∈Pmaxhte(γ) and MaxSp(M,P)=maxhtP(T1M). Remark. LetC be the set of all cusps of M . Under the same hypotheses as in Corol-lary5.4, the following two assertions hold, by applying Corollary5.3to the family of horoballs (gH′e′)g∈Γ/Γξ
e′ ,e′∈C with He′′ =He if e′ =e, and He′′ =He′[t] for some t big enough otherwise, for the first assertion, and to the family (gHe[tP])g∈Γ/Γξe ,e∈P
for the second one.
(1) For every cusp e of M , there exists a constant t≥0 such that for every h≥t, there exists a locally geodesic lineγ in M such that maxhte(γ) =h and maxhte′(γ)≤t for every cusp e′ 6=e in M .
(2) LetP be a nonempty set of cusps of M . Then MaxSp(M,P) contains the halfline [c′′2(∞)/2+tP,+∞].
Now, we prove the analogs of Corollaries5.3and5.4for families of balls with disjoint interiors. Let
Rmin0 =7 sinh c′1(∞)+3
2c′1(∞)≈22.4431.
Corollary 5.5 Let X be a complete simply connected Riemannian manifold with sectional curvature at most−1and dimension at least 3, and let Bn
n∈N be a family of balls in X with disjoint interiors such that the radius R0 of B0 is at least Rmin0 . For every h∈
c′′1(Rmin0 ,0,0),2R0−2 c′1(Rmin0 )
, there exists a geodesic line γ in X with phB0(γ)=hand phBn(γ)≤c′′2(Rmin0 ) for all n≥1.
Proof. We start by some computations. Letǫ >0. With c1=c′1(ǫ) and c5 =c5(ǫ,0) as in Subsection4.2, we have c5≥6 sinh c1 since c′3(ǫ)≥3 by the definition of c′3(ǫ) in Equation (- 6 -). By the definition of h0 in Subsection 4.2 and of h′ in Equation (- 5 -), we have h0(ǫ,0,c′1(∞)) ≥h′(ǫ,sinh c1) ≥2 sinh c1. Hence, by the definitions of c′′2(ǫ),c′′1(ǫ,0,0),h′1 and as ǫ7→c′1(ǫ) is decreasing,
c′′2(ǫ) =c′′1(ǫ,0,0)+c′1(∞)+2 c1
= max{2 c1,h0(ǫ,0,c′1(∞))+2 c5}+c′1(∞)+2 c1
≥2 sinh c1+12 sinh c1+c′1(∞)+2 c1 ≥14 sinh c′1(∞)+3 c′1(∞). Define nowǫ=Rmin0 , so that 2ǫ≤c′′2(ǫ) and R0 ≥ǫ. For every n6=0, let Rn be the radius of the ball Bn. If for some n 6=0 we have 2Rn ≤c′′2(ǫ), then phBn(γ) ≤c′′2(ǫ) and the last assertion of Corollary5.5 holds for this n. Hence up to removing balls, we may assume that Rn ≥ c′′2(ǫ)/2 ≥ ǫfor every n 6=0, so that the balls in (Bn)n∈N
areǫ-convex.
The end of the proof is now exactly as the proof of Corollary5.3, with the following modifications: ǫ = Rmin0 ; c′′1 = c′′1(ǫ,0,0); c′′2 = c′′2(ǫ); ξ is any point in ∂∞X ; Cn = Bn for every n in N; we apply Theorem 5.2 instead of Theorem 5.1, which is possible by the range assumption on h; we take now c1 = c′1(ǫ), so that we still have d(x′,y) ≤ c1 by Proposition 2.3; ξn is now the center of Bn, and βξn(u,v) = d(u, ξn)−d(v, ξn) (see Section2.1). Besides that, the proof is unchanged.
A heavy computation shows that
c′1(Rmin0 )≈1.7627,c′′1(Rmin0 ,0,0)≈101.4169 and c′′2(Rmin0 )≈106.7051 . Note that the above corollary is nonempty only if R0 ≥c′′1(Rmin0 ,0,0)/2+c′1(Rmin0 ) ≈ 52.4712. The constants in the following corollary are not optimal. Theorem1.3in the introduction follows from it.
Corollary 5.6 Let M be a complete Riemannian manifold with sectional curvature at most−1 and dimension at least 3, let (xi)i∈I be a finite or countable family of points in M with ri =injMxi, such that d(xi,xj)≥ri+rj if i6=jand such that ri0 ≥56 for some i0 ∈I. Then, for every d ∈[2,ri0 −54], there exists a locally geodesic line γ passing at distance exactly d from xi0 at time 0, remaining at distance greater than d from xi0 at any nonzero time, and at distance at least ri−56 from xi for every i6=i0. In particular,
mint∈Rd(γ(t),xi0)=d.
Proof. Let π : Me → M be a universal covering of M , with covering group Γ, and fix a liftexi of xi for every i ∈I . Let Bi be the ball BMe(exi,ri). Apply Corollary5.5to the family of balls (g Bi)g∈Γ,i∈I in X =M , which have pairwise disjoint interiors bye the definition of ri and the assumption on d(xi,xj). Note that ri0 ≥ 56 ≥ Rmin0 (see the definition of Rmin0 ). Let h =2(ri0 −d), which belongs to [108,2ri0 −4], which is contained in [c′′1(Rmin0 ,0,0),2ri0 −2 c′1(Rmin0 )] by the previous computations. Then Corollary5.5implies that there exists a geodesic line eγ in M such thate phBi0(eγ) =h and phgBi(eγ) ≤ c′′2(Rmin0 ) < 108 for all (g,i) 6= (1,i0). Parametrize eγ such that its closest point to exi0 is at time t = 0. Let γ = π◦eγ, then the result follows by the
definition ofphC (see Subsection3.1).